\documentclass[12pt, letterpaper]{article}

%==============Packages & Commands=================
\usepackage[utf8]{inputenc}
\usepackage{graphicx} % Graphics
\usepackage{indentfirst} % Tells LaTeX to indent every paragraph 
\usepackage{setspace} % To set line spacing
\usepackage[longnamesfirst]{natbib} % For references
\usepackage{booktabs} % For tables
\usepackage{rotating} % For sideways tables/figures
\usepackage{amsmath} % Some math symbols
\usepackage[margin = 1in]{geometry}
\usepackage{subcaption}
\usepackage{mdwlist}
\usepackage{pifont}
\usepackage{url}
\usepackage{verbatim}
\usepackage{textcomp}
\urlstyle{same}
\usepackage{appendix}
\usepackage{multirow}
\usepackage{dcolumn}
\usepackage{threeparttable}
%\usepackage[nolists]{endfloat} % Figures and tables at the end
\usepackage{cleveref}
\usepackage{kotex}
\usepackage{tikzit}
\input{dag_styles.tikzstyles}
\usepackage{threeparttable}
\usepackage{authblk}


%These allow for references to subfigures


\newcommand{\R}{\texttt{R}\space} % Write R in typewriter font
\newcommand{\trans}[1]{{#1}^{\ensuremath{\mathsf{T}}}} % Transpose symbol 
\bibpunct{(}{)}{;}{a}{}{,} % Reference punctuation 
\captionsetup[subfloat]{position = top, font = normalsize} % For sub-figure captions
\graphicspath{{Figures/}}
\DeclareGraphicsExtensions{.pdf, .png, .jpg, .tif, .tiff, .jpeg}

%\DeclareDelayedFloatFlavor{sidewaystable}{table} %Allows endfloat to handle sidewaystable
%\DeclareDelayedFloatFlavor{sidewaysfigure}{figure} %Allows endfloat to handle sidewaysfigure

%These packages for inserting blue hyperlinks in text (and for linking in-text references to references section)
\usepackage[bookmarks]{hyperref} %bookmarks option enables pdf bookmarks to be automatically generated
%\usepackage[usenames,dvipsnames,svgnames,table]{xcolor}
\hypersetup{colorlinks=true,
linkcolor=blue,
citecolor=blue,
filecolor=blue,
urlcolor=blue,
bookmarksnumbered=true,     
    bookmarksopen=true,         
    bookmarksopenlevel=1} % Link colors

% Need tikz to draw causal path diagrams
\usepackage{tikz-cd}
\usepackage{tikz}

\usetikzlibrary{positioning, calc}

\def\pgf@matrix@xscale{1}
\def\pgf@matrix@yscale{1}
\makeatother

%For nice-looking quotes:
\usepackage{epigraph}
% \epigraphsize{\small}% Default
\setlength\epigraphwidth{16cm}
\setlength\epigraphrule{0pt}

\usepackage{etoolbox}

\makeatletter
\patchcmd{\epigraph}{\@epitext{#1}}{\itshape\@epitext{#1}}{}{}
\makeatother

%\usepackage{authblk}

\usepackage{float}
\usepackage[multiple]{footmisc}
\usepackage{multirow,array}
\makeatletter
\usepackage{longtable}
\setlength{\@fptop}{0pt}
\makeatother
%===============Some new commands==================
\newcommand{\fnote}[1]{\footnote{\begin{doublespace}\normalsize{#1}\vspace{-26pt}\end{doublespace}}} % 12 pt, double spaced footnotes 
%\renewcommand{\figureplace}{ % This places [Insert Table X here] and [Insert Figure Y here] in the text
%\begin{center}
%[Insert \figurename~\thepostfig\ here]
%\end{center}}
%\renewcommand{\tableplace}{%
%\begin{center}
%[Insert \tablename~\theposttbl\ here]
%\end{center}}

%============Article Title, Authors================
%\title{Why Political Connections Can Impede Investment}
%\author{Robert Kubinec, Haillie Lee, and Andrey Tomashevskiy}
%\date{\today}

\title{Why Corporate Political Connections Can Impede Investment}

\author{}

%\author[1]{Robert Kubinec\thanks{Corresponding author: \texttt{rmk7@nyu.edu}}}
%\author[2]{Haillie N. Lee}
%\author[3]{Andrey Tomashevskiy}

%\affil[1]{Division of Social Sciences, New York University Abu Dhabi}
%\affil[2]{Department of Political Science and International Relations, Seoul National University}
%\affil[3]{Department of Political Science, Rutgers University}
\date{\today}

%===================Startup========================
\begin{document} 
\maketitle
%\doublespacing
%\vspace{5.5in}
%\noindent Note: Note on replication data.
\thispagestyle{empty}
%===================Abstract======================= 

%IO does not want the abstract, word count, or acknowledgments submitted in the manuscript

%\newpage
\thispagestyle{empty}
\begin{abstract}
\begin{normalsize}
\noindent 

We present an experiment that manipulates corporate political connections to understand whether a company's political influence is a barrier or an inducement to investment. Our data come from a survey of 3,329 firm employees and managers located in Venezuela, Ukraine and Egypt. On the whole we find that our respondents do not prefer to invest in companies with political connections. These results, however, are highly conditional on the respondent's company: respondents from highly connected companies do not penalize connected companies as investment choices, while respondents at less-connected companies prefer to invest in companies without political connections. We believe that what explains this finding are differences in how companies with and without connections manage liability as our survey data shows connected companies are much more likely to employ informal rather than formal mechanisms to resolve disputes. As a result, we believe that unconnected investors are more likely to invest in other unconnected companies to ensure that their property rights are protected.

%\footnote{Code and data to reproduce analyses is available via the Github repository: \url{https://github.com/saudiwin/exprop_survey_public}. We thank participants of the 2020 International Political Economy Society Conference, participants of the 2020 American Political Science Association Annual Conference and Jong Hee Park and Sung in Kim for helpful comments. We thank the Niehaus Center for Globalization and Governance at Princeton University for providing funding for this project. We also thank the Division of Social Sciences at New York University Abu Dhabi and Rutgers University for funding and support.}

%\footnote{An anonymized version of this article's pre-registration is available in the supplemental information.}


% I'M SAVING OLD ABSTRACT BEFORE REVISING IT - BOB
%A growing body of research in political science and economics examines the causes and consequences of expropriation of foreign direct investment. While this research often focuses on government-level decisions to expropriate, firm-level determinants of expropriation remain unexamined: why are some firms targeted for expropriation and government predation while other firms are left unharmed? We attempt to answer this question with a unique survey of firms in Venezuela, Ukraine and Egypt. We argue that a firm's political connections determine the likelihood of expropriation. Using firm-level evidence, we show that expropriation risk is not uniform across all firms in an investment host country: firms lacking political connections and a credible threat of exit face higher risks compared to well-connected firms and those with viable exit options. Additionally, we find that not all political connections are equal. Political ties to highly ranked or incumbent political actors are more valuable. Lastly, we show that firms use different countermeasures to deal with government predation based on the degree of political connections that they enjoy. While firms with strong political connections are more likely to leverage their ties to combat government predation, firms with weak or no political connections are likely to resort to formal measures such as domestic lawsuit or international arbitration. 


%We find that political connections moderate the effect of firms’ characteristics on expropriation risk and reduce political risk for firms that would otherwise be vulnerable to government predation. 

\end{normalsize}
\end{abstract}


%================Begin Manuscript==================
\newpage
\doublespacing

\section{Competing Interests}

The authors have no competing interests to declare.

\section{Introduction}
\label{sec:intro}

Firms operating in high-corruption countries with weak property right often rely on informal connections to political agents to succeed in an environment fraught with risks \citep{fisman2001estimating,faccio2006political}. These types of non-market strategies can be effective in securing preferential access to loans and contracts, as well as providing protection from predation by government agents. A large body of work shows that political connections are an important aspect of business-government relations and firms' non-market strategies in countries with weak institutions \citep{shleifer1993corruption,samphantharak2008predictable}, but at the same time, corruption writ large is a hindrance to investment and innovation \citep{mauroEconomicIssuesNo}. In this paper, we examine to what extent political connections help companies obtain domestic investment. To do so, we focus on investment by business investors within a country, which can take the form of individual, familial or interfirm (intercorporate) investment and represents a sizable share of investment activity across countries \citep{fedeniaCrossHoldingsEstimationIssues1994}.

Focusing on domestic sources of capital is very important to understand how companies survive within environments in which property rights protection is highly variable \citep{markus2015property,frye2004credible}. For these reasons, politically connected firms should be more attractive options for investment because domestic investors can benefit indirectly from these connections. The wealth of studies documenting above-market returns to political connections shows that political connections can be a very valuable asset to companies, especially in developing countries \citep{fisman2001estimating,acemoglu_power_2017,earle2015}. 

However, it is possible that connected firms could be unattractive investment options since such firms may be more likely to break contracts and engage in opportunistic behavior \citep{Williamson_1979,bandari2020}. Connected firms may also have the ``wrong'' type of connections to political agents out of favor with the ruling regime in a given country \citep{albertus2012if,earle2015,frye2016elections}. Cooperation is less attractive if firms with the ``wrong'' connections, along with their partner firms, face more expropriation and government predation. Although existing research suggests that political connections are a valuable asset, it is not clear if and how much outside investors would value political relationships when considering whether to invest in a connected company.

% we should go over the argument logic in the theory/background section

%We focus on domestic firms and argue that firms prefer to cooperate with other firms that have similar levels of political influence. We believe that part of what explains this homophily is the availability to companies of formal vs. informal means of redress for contractual and criminal malfeasance \citep{Gans-Morse_2017,Marques_Levina_Kazun_Yakovlev_2020,Markus_2012}. Unconnected firms are more likely to use formal means of dispute resolution, such as courts, to address conflicts with business partners. In high-corruption countries, these institutions are likely to be weak and ineffective. Connected firms, on the other hand, are more likely to rely on informal means by leveraging their connections to government agents. These informal methods, involving extralegal economic exchanges or direct interventions by powerful government officials, may be more effective in resolving disputes and generating favorable outcomes for connected firms.

% trying to shorten the introduction, we address thislater in the background

%Although cooperation between connected and unconnected firms may be useful for unconnected firms hoping to benefit from their partner's connections, this type of cooperation is also more risky. Since unconnected firms lack access to informal means of redress, a mismatch in business-government relations creates a power imbalance. Connected firms may engage in opportunistic behavior costly to business partners with impunity since connected firms may rely on connections and informal dispute resolution to escape penalties and ensure a favorable result \citep{Qian_Pan_Yeung_2010}. Forward-looking unconnected firms may anticipate this and would thus be less likely to cooperate with connected firms. Firms are therefore likely to cooperate with firms that have similar relations with government agents and access to similar means of redress: unconnected firms are more likely to cooperate with other unconnected firms, while connected firms are more likely to cooperate with other connected firms.

A central problem in evaluating this research question is that political connections are almost never randomly assigned to companies and thus analyses with cross-sectional or even panel data can be obscured by selection bias. To address these inferential concerns, we conducted an online survey experiment by using Facebook advertisements to recruit 3,329 business employee and manager respondents in Egypt, Ukraine and Venezuela. Survey respondents were asked to describe their firm's characteristics, degree of political connections, and experience with government predation and expropriation. 

%These data represent a unique source of information on the relationship between political connections, firm characteristics, and expropriation. By using Facebook advertisements, we are able to gather data in countries where traditional surveys are more difficult to conduct. We are thus able to generate valuable firm-level insights regarding the relationship between political connections and inter-firm cooperation.

We implemented an investment choice experiment in the survey that allows us to experimentally manipulate company political connections along with other company-level variables such as profitability, size and country of origin. Respondents were presented with characteristics of two hypothetical firms and asked to choose how much money to invest in each. On the whole, respondents were \emph{less} likely to choose companies with political connections. Examination of treatment heterogeneity showed that this effect is driven by the experiences of respondents at their companies. Respondents at companies with more political connections (and specifically more \emph{valuable} political connections) prefer to invest in connected companies, while respondents at non-connected companies prefer to invest in non-connected businesses. We also show that companies' ability to use informal forms of redress is also a moderator of the treatment, suggesting that the reason political connected respondents prefer politically-connected investments is due to the need to have equal standing when addressing contractual disputes.

%While other research has focused on factors such as trust, and cultural distance to explain interaction among firms \citep{Farrell_2005,Nielsen_2007}, we show that a firm's political strategy plays an important role in affecting that firm's attractiveness as a business partner. 

This paper makes contributions to research on political connections and business-government relations by experimentally evaluating the role of political connections with respect to investment. The highly conditional effects of connections may help explain why in high-corruption countries private domestic investment tends to be lower on average, depressing long-run economic growth \citep{tanziCorruptionPublicInvestment1998}. Our theory and results suggest that investors operating in high-corruption environments must consider not only the risk of expropriation and opportunism by government agents but also risks associated with their business partners \citep{Henisz_2000,Qian_Pan_Yeung_2010,Markus_2012}. This contributes to a large literature on property rights, expropriation, and political risk in political science, economics and management that has largely focused on acts of expropriation by government agents \citep{kobrin1987testing,jensen2003democratic,jensen2008political,li2003reversal,li2009democracy}. In adopting a cross-national approach, we are also able to show that the role of political connections with respect to investment is relatively similar across countries with differing political systems yet similar issues with corruption. 


\section{Theory}
\label{sec:intro}

A large literature on business-government relations points to the importance of political connections as a component of firms' so-called ``non-market" strategies. Prior research shows that firms with strong political connections receive significant preferential treatment (e.g. better access to credit market and contracts) from governments, especially if these governments are known to struggle with accountability and transparency. Political connections involve direct ties between a firm and government structures \citep{haber_politics_2003,frye_property_2017,markus2015}. Pioneering studies showed that politically-connected firms were on average more valuable than those without such connections, which was especially evident during moments of transition when these connections evaporated \citep{roberts1990dead, fisman2001estimating,earle2015,acemoglu_power_2017}. In another example, \citet{faccio2006political} found that politically connected firms are more likely to be bailed out relative to comparable non-connected firms. Similarly, \citet{claessens2008political} showed that firms with political connections enjoy better access to bank financing, which is a critical constraint in developing economies with limited access to capital markets.

In high-corruption environments where government predation is common, officials may create exclusive ``spheres of influence'' that correspond to formal and informal hierarchical relationships. The presence of these ties implies that officials hold some stake in the survival of the firm and are willing to intervene on behalf of the firm. Given this, officials may be more reluctant to target firms that have ties to other, potentially more influential, government officials \citep{fried2010corruption}. Political ties can work as insurance and protection from uncontrolled and unpredictable predation if some subset of officials, who would otherwise target a firm, are deterred from interfering with its operations \citep{keillor2005threats, campos1999impact}. Firms with political connections may thus incur fewer political risks and will be able to operate more efficiently compared to politically unconnected firms \citep{galang2012victim}.\footnote{This does not necessarily imply that politically connected firms incur no costs. In building ongoing relationships with government officials, politically connected firms may simply be transforming the unpredictable costs of corruption (e.g. unanticipated and intermittent incidents of expropriation) into a stabler stream of ongoing costs to nurture the political connections (e.g.bribes) \citep{wei1997corruption}.} 

These benefits of political connections mean that to companies, political connections should be a type of asset, and thus, ceteris paribus, companies with these connections should be more attractive targets for investment. Political connections may help with protecting contractual claims via courts \citep{ang_perverse_2014}, purchasing land \citep{10.1093/qje/qjy027}, obtaining permissions to enter regulated sectors \citep{diwan_pyramid_2015}, and avoiding costly inspections and tax audits \citep{tsai_collusion_nodate}. 

However, recent research raises important questions about whether political connections are always beneficial for companies.  \citet{bandari2020} and \citet{bhandari_social_2022} showed that political connections deterred prospective buyers of mobile credit and also, to some extent, managers from procuring supplies from connected companies. As both of these studies are experimental in nature, whereas most of the extant research on political connections is observational, these contrasting findings suggest that the estimated advantage accruing to political connections may in part be based on selection or collider bias in which we only observe a subset of the possible outcomes for companies and connections. In another vein, \citet{diwan_political_2020} found that politically-connected firms in Lebanon over-employed people in order to provide clientelist perks to constituents, a dynamic which has also been observed in France \citep{bertrand_cost_2018}, suggesting that political connections can also be a liability for firm performance. 



%A simple reading of the literature would suggest that political connections are nearly always beneficial except in the rare case of a government collapse, and thus all companies should seek to have them. Nearly all of the studies, though, implicitly examine direct effects of political connections by comparing connected to non-connected firms and assuming that the two classes of companies are functionally independent of each other. In this study, we take a broader lens to examine how political connections affect relationships between companies and between companies and investors, which will have significant effects on economic development over time. 

%How do political connections affect domestic investment? Existing research does not give a direct answer, though we can infer one based on previous findings. In general, it would seem that political connections should provide important benefits for investors and business partners. Firms cooperate by entering into strategic alliances such as joint ventures or equity alliances in order to leverage resource complementarities and improve market share. Political factors may also drive incentives to cooperate if strategic alliances reduce risks of expropriation and predation by government agents. Firms may thus enter into joint ventures with foreign or domestic firms to increase protection of property rights \citep{Henisz_2000,Betz_Pond_2019}. Structural institutional factors, as well as firm-level characteristics, may reduce transaction costs and increase the likelihood and efficiency of inter-firm cooperation \citep{Farrell_2005, Nielsen_2007}.



%Political connections are likely to play an important role if cooperation with politically connected firms provides access to some of the benefits associated with political connections. Existing research suggests that firms should be more likely to cooperate with politically connected firms to protect property rights and reduce the risks of expropriation. 

The main reason we believe that connections could be a problem for domestic investment is because strategic alliances involving connected firms and unconnected investors imply a power imbalance. Since unconnected investors lack informal ties to government agents, unconnected investors must rely on formal dispute resolution mechanisms in the event of conflict with connected firms \citep{Gans-Morse_2017,Marques_Levina_Kazun_Yakovlev_2020,Markus_2012}. Connected firms are more likely to act opportunistically when interacting with unconnected investors since they can leverage their ties to government officials, bypass formal institutions such as courts, and escape formal penalties. In the event of a dispute, connected firms are more likely to leverage the power imbalance and receive a favorable outcome due to their connections to government agents. 

While the existing literature largely considers threats to property rights to originate outside the firm (such as government action), the power imbalance between connected firms and unconnected investors implies that investors face additional risks from within the firm. Politically connected firms may engage in more opportunistic behavior if political connections shield connected firms from costly punishment \citep{bandari2020}. In a high-corruption environment, connected firms are more likely to rely on various informal means of dispute resolution instead of formal institutions, such as courts or law enforcement authorities. Freed from formal institutional constraints, connected firms may interfere with the property rights of investors and business partners by misallocating profits, altering ownership share, or directly taking over investor property. \citet{leuz_political_2006} identify a similar dynamic in firm financing, where connected firms are less likely to seek foreign financing due to a predilection for opportunistic behavior and a desire to minimize transparency. Unconnected investors face a higher degree of vulnerability and risk when investing in a connected firm. While investment in a connected firm may shield investors from predation by government agents, it may instead create new risks of predation by business partners.\footnote{The type of dynamic described here identifiable in each of the three countries included in our survey experiment. one particularly instructive example involves conflicts in 2020 surrounding Citrus, the largest electronics retailer in Ukraine, valued at \$68 million. Grigoriy Topal, a joint owner of Citrus, attempted to dispossess his unconnected business partner Gennadiy Zinchenko of his shares in the company through fraudulent means. Topal relied on his political connections to law enforcement authorities in the Office of the Prosecutor General to escape criminal responsibility and was ultimately successful in taking full control of the company through extra-legal techniques.(https://kp.ua/economics/688772-peredel-tsytrusa-hryhoryi-topal-pytaetsia-otobrat-torhovye-marky-na-myllyony-dollarov). In 2022, Venezuela exhibited a similar scenario. Petróleos de Venezuela S.A.(PDVSA), the Venezuelan state oil company, abruptly seized a minority stake in a crucial joint oil venture located in Venezuela from GPB Global Resources, a private energy firm established by ex-officials of Gazprom. This sudden move was coupled with an abrupt overhaul of the joint venture's board of directors, an action seemingly communicated to GPB only through a brief notification letter. The legitimacy of these actions is questionable as there's no clear indication of GPB's consent, thus highlighting the predatory nature of the conduct by the politically connected PDVSA. 
(https://www.bloomberg.com/news/articles/2022-09-30/venezuela-seizes-stake-in-oil-venture-from-firm-with-russia-ties#xj4y7vzkg). 
Lastly, an ownership dispute in Egypt between Orascom Telecom, a local entity, and France Telecom, a multinational corporation, over Mobinil, an Egyptian mobile operator, offers a clear depiction of how local political connections can exert influence on regulatory decisions and disadvantage corporate entities lacking such connections.  Despite a ruling from the International Court of Arbitration in favor of France Telecom, which directed it to purchase Orascom's 28.75\% stake in the holding company at US\$48.80 per share, Egypt's market regulator, the Capital Market Authority (CMA), rebuffed France Telecom's bids for the remaining shares on two occasions. In the wake of these obstacles, France Telecom publicly accused the CMA of exhibiting "double standards," tacitly suggesting an abuse of Orascom's domestic political sway.(https://developingtelecoms.com/telecom-business/telecom-investment-mergers/1831-france-telecom-orascom-dispute-over-mobinil-erupts.html)}

Cooperation with politically connected firms thus carries additional risks internal to the firm; transaction costs, risk of hold-up and opportunism can generate new uncertainties and costs for cooperating firms. Unconnected investors must then weigh the risks of opportunistic behavior when evaluating prospective partners. Risks of costly opportunistic behavior are likely higher when involving firms with connections to more powerful officials, implying fewer constraints. This type of risk is likely to be lower for pairs of unconnected companies and investors since both sides lack political connections and would need to rely on formal institutions with a more level playing field in event of a dispute. This logic implies that unconnected investors are less likely to invest in connected firms and are more likely to invest with other unconnected firms. While connected firms may actually prefer to cooperate with unconnected firms, connected firms will be less likely to locate willing unconnected partners and will then cooperate with other connected firms. If both sides of the investment relationship have political connections, both have access to informal as well as formal means of dispute resolution.

%Furthermore, firms may have access to the ``wrong'' connections to governments agents out of power. Officials that are in office currently are more likely to exert influence on behalf of a particular firms, compared to former officials. If there has been a recent regime transition, association with former officials may actually be hazardous \citep{siegel_contingent_2007,sun_dynamic_2010,markus_flexible_2017}. \citet{albertus2012if} note that a new dictator may expropriate firms associated with preexisting elites to signal his loyalty to the organization that has launched him into power. \citet{frye2016elections} find that connections to the ruling party matters. Firms with close economic ties to the state felt more vulnerable to government expropriation after a surprising parliamentary election that weakened the bargaining power of the ruling party. When choosing targets for investment, investors with the ``wrong'' connections may thus also choose to cooperate with unconnected firms, rather than firms that benefit from their current connections. The imbalance in power between investors and firms is thus likely to persist among connected firms if some investors have ineffective connections to officials that are out of power.

Finally, investors may face additional risks due to a lack of transparency and ineffective oversight. If connected firms leverage connections to bypass institutional constraints, connected firms are less likely to provide transparent and accurate information \citep{bona-sanchez_politically_2014}. Connected firms may also be less likely to comply with safety regulation and more likely to engage in inefficient  or fraudulent financial practices. This create a higher risk of adverse outcomes for investors in connected firms. As an example, Ukraine's largest lender Privatbank was controlled by politically connected businessmen until 2016. In 2016, the bank experienced large capital losses and was nationalized by the Ukrainian government.\footnote{https://www.bbc.com/news/business-38365579} The capital losses that led to nationalization and loss of investor value resulted from fraudulent lending practices that were in part enabled by the owners' political connections. Since political connections may provide cover from law enforcement and regulatory authorities and remove constraints on illegal behavior, investmentment in connected firms may involve higher risks despite ostensibly high profitability.  

%While political connections can provide several benefits to firms, other studies suggest that not all political connections are equal. More specifically, benefits may vary with the political power of the government entity to which a firm is connected. \citet{khwaja2005lenders} find that political rents increase with the strength of the firm's connected politician and whether that politician is currently in power. Additionally, \citet{fisman2001estimating} shows that during episodes of rumors of Suharto's health in Indonesia in 1997, companies politically dependent on the government lost more stock market values on average compared to those with weaker connections. 

%Given this, we are interested in understanding which political connections prove to be the most valuable. It is likely that an owner's or board member's current ties to government structures are more valuable than historical ties. Officials that are in office currently are more likely to exert influence on behalf of a particular firms, compared to former officials. If there has been a recent regime transition, association with former officials may actually be hazardous. \citet{albertus2012if} note that a new dictator may expropriate firms associated with preexisting elites to signal his loyalty to the organization that has launched him into power. \citet{frye2016elections} find that connections to the ruling party matters. Firms with close economic ties to the state felt more vulnerable to government expropriation after a surprising parliamentary election that weakened the bargaining power of the ruling party. We expect that ties with the governing party will tend to be more valuable compared to former ties and connections to marginalized political actors with limited bargaining power. This insight leads to H2: 

%Moreover, it is likely that connections to high-ranking bureaucrats will be more valuable compared to those to lower-ranked officials. High-rank officials are more likely to have greater influence and the ability to work on behalf of a connected firm to limit expropriation and predation by other officials. For example, being politically connected to the powerful ruling party may bring more benefits compared to being connected to a smaller political party with less influence. These expectations led to H2: 

%interaction effects between political connections and other treatments. We have reason to believe that the value of political connections tends to increase with larger and more profitable firms. Bureaucrats may have a stronger incentive to target large firms that have more assets to expropriate. Larger firms thus stand to gain more by cultivating political connections. The negative effects of connections on expropriation should be more pronounced for larger firms compared to smaller firms.%



%The strength of a firm's political connection may also influence the type of countermeasures that it takes against government predation. Firms with strong political connections may leverage such connections to stop predation and receive remedies. For example, firms tied to the party in power may be able to avoid police inspections by pulling a few strings. On the other hand, firms with outdated or no political connections may have to resort to formal institutions such as domestic lawsuit or international arbitration to combat government predation. These expectations lead to H3 below:

%\begin{quote}
   % H3: While firms with strong political connections are likely to leverage their connections as a countermeasure to predation, firms with outdated or no political connections are likely to resort to formal measures such as domestic lawsuit or international arbitration.  
%\end{quote}

%To examine these hypotheses, we use data from a unique survey of firm employees. We describe the survey in the next section. These first three hypotheses we test with observational data collected on companies, but we also embedded an experiment within the survey. For our experimental data, we look at two additional hypotheses, H4 and H5, that were pre-registered prior to data collection. These hypotheses are specific to the experiment, which we describe later, that allows us to manipulate the political connections of companies by asking respondents to choose between two companies as investment opportunities.

Given this apparent divergence in the literature about the potential role and benefits of political connections with respect to investment, in this paper we examine how domestic investors' choices in developing countries are affected by political connections. To the extent that firm-government relations affect a firm's profits and earning potential, as well as the security of property rights, these are factors that should play a direct role when investors decide to put their capital into a given firm. 

We further aim to expand the study of political connections by adopting an expansive conceptualization of political connections that includes both formal and informal ties to government institutions and politicians while also employing experimental inference to learn about the effect of connections on investment. We define political ties as involving a CEO's, board member's, or manager's past or current appointments to positions in governmental institutions, but also personal relationships that involve company owners, managers, and board members. Alternatively, political connections may include membership in political parties, past or present positions in government, or direct family ties to presidents or prime ministers.

With this broad conceptualization, our experiment allows us to examine political connections in much greater detail than has previously been possible. To examine the relationship between investment and political connections, we test the following competing hypotheses:

\begin{quote}
    H1a:  Politically-connected companies should receive more investment than non-connected companies.
    
    H1b:  Politically-connected companies should receive less investment than non-connected companies.
\end{quote}

These hypothesis are based on the differing logics described above. Falsifying one of these hypotheses would help to resolve the ambiguous predictions based on existing research.  These hypotheses were pre-registered (see our pre-registration in the supplementary information) and are included here in a slightly re-worded form for clarity.\footnote{We note that while our pre-registration includes other hypotheses, for the purposes of this article we concentrate on these two. We hope to consider the other hypotheses in future research.}

In the supplemental information we discuss further the difficulty of identifying the effect of political connections on investment with observational data, and in particular how selection processes related to firm survival are likely to bias any effects. By using an experimental method, we can determine to what extent political connections have a direct effect on investment even if the relationship is usually obscured by selection biases related to the way that political connections help companies survive and enter particular sectors.

\section{Data}

To collect data for our conjoint experiment, we conducted an online survey using Facebook advertisements in three countries: Ukraine, Egypt and Venezuela. Our survey contains the experiment as well as questions that allow us to collect observational firm-level characteristics. We translated the survey to Arabic, Ukrainian and Spanish to ensure we could obtain a diverse sample. 

Ukraine, Egypt and Venezuela are appropriate countries for this type of analysis since these are high corruption countries where firms face political risks including demands for bribes, harassment, and shake-downs by government agents and a higher risk of expropriation. All of these countries are near the bottom of the 2019 Transparency International Corruption Perception Index: Ukraine ranks 126/198, Egypt is 106/198 while Venezuela is 173/198. However, Venezuela, Egypt and Ukraine have very different institutions: Venezuela and Egypt are authoritarian regimes while Ukraine is a democratizing country with a competitive semi-authoritarian past where elections are generally recognized to be free and fair.\footnote{Data for all countries collected in the winter and summer of 2020.} Our country selection for this reason follows a most-different systems design in which we identify countries that are similar in terms of our independent variable of interest--corruption and crony business-state relations--while maximizing diversity in terms of ancillary factors that could affect investment such as the type of political institutions. By so doing we hope to be able to see if political connections tend to have similar effects across diverse institutions, increasing the external validity of the analysis.

Our survey targeted Facebook ads at individuals 18-64 who own or are currently employed across all firms in Ukraine, Egypt and Venezuela. By doing so, we leverage the power of Facebook to target subsets of the population \citep{jager_potential_2017}. While Facebook surveys may not return a balanced nationally-representative samples, the extraordinary size of its user base at nearly 3 billion monthly users means Facebook samples are often quite close to even ``gold standard" household surveys \citep{rosenzweig_survey_2020}. Furthermore, our aim is to recruit from a distinct sub-population that cannot be easily targeted via traditional methods such as household recruitment. Given that we are asking about company data that can be sensitive, such as political connections, the ability to ask these questions in a secure and anonymous online setting is a major advantage over traditional in-person interviewing methods. Because Facebook has a strong financial interest in identifying business managers among its users for the purpose of business to business marketing, we can employ their ad targeting to obtain responses from this population of companies without needing to know the companies' individual identities. 

While online samples have limitations with respect to population inference, it is important to note that our interest is in estimating an average treatment effect (ATE), not population totals. Online surveys may have imbalanced panels yet reproduce ATEs from nationally representative surveys \citep{coppock_generalizing_2019} because people respond to the treatment similarly across demographic strata. For these reasons, we employ best practice in Facebook sampling by using ads targeted at diverse sample characteristics and then check for possible ATE heterogeneity across these characteristics \citep{neundorf_recruiting_2021}. As our population is employees of firms, we naturally focus on the company's sector as the most likely source of heterogeneity in ATEs.

To ensure sectoral diversity, we employed Facebook profile ad data to separately target ads at managers and employees in distinct sectors, such as manufacturing, retail, finance, services and information technology, in each country separately, for a total of over thirty distinct advertisements. We collected responses from 6,076 individuals, of whom 3,329 completed the experiment, including employees, managers and CEOs of companies in the target countries. 1,005 of the completed responses came from Egypt, 1,512 from Ukraine, and 812 from Venezuela. These individuals represent firms varying across sectors, income, number of employees, and includes both domestic as well as wholly owned subsidiaries and joint ventures of foreign firms. Our sample includes 59.08\% employee respondents and 40.92\% management respondents based on self-reporting in the survey.

\begin{figure}[!h]
\caption{Sample Sector Distribution}
	\centering
		\includegraphics[scale=.50]{sectors2.png}
	\label{fig:sector}
		\scriptsize
	Note: This figure includes additional respondents who answered this question in the survey but did not also complete the experimental treatments.
\end{figure}

\begin{figure}[!h]
\caption{Sample Firm Size Distribution}
	\centering
		\includegraphics[scale=.30]{size2.png}
	\label{fig:size}
	\scriptsize\\
	Note: This figure includes additional respondents who answered this question in the survey but did not also complete the experimental treatments.
\end{figure}

\begin{figure}[!h]
\caption{Ukraine Sample and Census Sector Distribution}
	\centering
		\includegraphics[scale=.7]{samplecompar3.png}
	\label{fig:compare}
\end{figure}

We present some descriptive information on the distribution of firms across sectors and sizes in Figures \ref{fig:sector} and \ref{fig:size}. We use NACE two-digit sector codes to classify firm sectors. Figures \ref{fig:sector} and \ref{fig:size} demonstrate that our data contain information on firms of of varying sizes operating across a wide variety of sectors. To compare how our sampled data compared to the observed distribution of firms, we use sector data from the Ukrainian firm census. While we are unable to use a similar census to examine the representativeness of our Venezuelan or Egyptian sample, the difference between firm proportions in our sample vs. the Ukrainian firm census is still informative. We present these differences in Figure \ref{fig:compare} and include 95\% confidence intervals to account for the uncertainty inherent in the sampling process. 

While our data over-samples educational firms and under-samples retail firms, we do see substantial representation across sectors, which is evidence that our sampling strategy was able to recruit a reasonably diverse sample. While we cannot make formal claims of representativeness, we note that obtaining a random sample of companies is much more difficult than with individuals due to the considerable challenges to defining the sampling frame and dealing with nonresponse bias. As such, given the constraints of business sampling in developing countries and our ability to recruit a reasonably diverse sample, we believe that these findings are informative of a broad swath of the business community in these countries. Furthermore, we are able to check for treatment heterogeneity across sampling characteristics to know how much imbalances in the sampling frame may affect inference to the population.

Respondents to the survey were asked to answer a variety of questions soliciting information on firm characteristics, experiences with governments involving direct occurrences and interactions, as well as various opinions and perceptions. 

To measure political connections, we use information from two variables. First, we asked respondents to describe their firm's level of political connections on a 10-point scale, ranging from unconnected at 0, to strong political connections at 10. We also asked respondents to describe the efficacy of their connections on a 10 point scale. This variable ranges from ``hurt by connections'' at 0, to ``benefits from connections'' at 10.\footnote{The Appendix contains the questions used in these analyses.} This formulation is perceptual: we rely on the respondent's idiosyncratic estimation and interpretation of the firm's political connections and may include difficult-to-quantify informal relationships between the company and power brokers. 

% Second, we asked respondents if their firm has a manager, CEO, owner or board member who are current or former members of parliament, high-ranking or low-ranking bureaucrats. We use a count of the total number of \textit{high-ranking officials and parliament members} that are associated with a given firm to measure these formal ties. Firms are likely to have strong political ties if multiple high-ranking officials or parliament members have a stake in ensuring the continued success of a firm and have an incentive to use their influence on behalf of the firm. 

\begin{figure}[!h]
\caption{Political Connections and Efficacy}
	\centering
		\includegraphics[scale=.15]{pol_con_eff.png}
	\label{fig:pol_con_eff}
\end{figure}

We present some descriptive results about our two 1 to 10 scales of connections in Figure \ref{fig:pol_con_eff}.  For companies with no connections, they may think that this level of connections is perfectly fine (a high efficacy) or not enough (a low efficacy). Conversely, for connected companies, they may think their connections are benefiting the company (high efficacy) or are in fact a liability to the company (low efficacy). Figure \ref{fig:pol_con_eff} plots the relationship between these variables by color-coding each cell with the count of respondents reporting a given level of connectedness and the efficacy of those connections. While the two variables are highly correlated (correlation coefficient of .6), suggesting that firms tend to benefit from political connections, we observe a considerable amount of off-diagonal cases as well. Some highly connected firms thus report that they are hurt by their connections, suggesting that effects of political connections are not uniform. Similarly, while some unconnected firms would prefer to have higher level of connections (low connection, low efficacy), others may prefer to avoid politics and are satisfied with their connection levels, or lack thereof (low connection, high efficacy). We use the 10 point connection level scale in the analyses described below.

\subsection{Experimental Design}

Our experimental treatment necessarily abstracts away from details while still preserving as much realism as possible about domestic investment in developing countries. Abstraction is necessary to permit generalizability across countries and individuals, and we note that recent research has shown that abstraction, including the employment of hypotheticals, does not necessarily hurt construct validity \citep{brutger_abstraction_nodate}. As such, we seek to balance both realism and abstraction in the design by presenting a relatively simple, small-scale investment scenario that would be widely plausible in developing countries yet still incorporates multiple investment attributes. 

Our design is based around the fact that raising capital from business colleagues, friends and family is a common way to capitalize businesses in developing countries where banks are unable to meet financial needs and political connections or other types of elite status may determine who gets loans \citep{autio_economic_2015,raza_institutional_2020,batjargal_institutional_2012,ge_who_2019}. Given what we know about how companies are often financed in these countries through informal networks, we think that the scenario presented to these respondents is plausible as it likely that the respondents may have been approached by business associates in the past asking them to help finance a new or existing venture. 

%Sample representativeness can be difficult to judge because our sampling frame is not aimed at creating a nationally representative sample, but rather a very specific subset of the population. As such, traditional re-weighting methods based on census demographic data are not applicable in this context. Indeed, the social media recruitment strategy helps us to target much more specific demographics than we could with traditional survey methods.

%However, we do note that prior research has shown that online RCTs, even those from very limited populations such as MTurk, in fact replicate quite well in national samples \citep{coppock_generalizing_2019}. Because we can randomly assign the treatment, we know that sample characteristics will only affect our estimand to the extent that there is treatment heterogeneity. As such, while explicitly correcting for sample composition would be quite difficult, by exploring subgroup analysis as we do in this paper, we can learn how the treatment effect might change if sample composition changed, which can help us learn how these findings would travel to other samples and settings.

%To measure political predation, which is our key dependent variable, we ask respondents to provide information on their experience in interacting with government officials. First, we ask respondents to report if government officials have confiscated either property/assets, income/profits or both without fair compensation (\textit{Confiscation}) over the past two years. Second, we use the total number of inspections by tax agencies, police/law enforcement, safety officials, and health officials (\textit{Inspections}) over the past year. These variables represent various types of predatory activities that may be pursued by government officials to pressure firms and extract value.

%To account for the heterogeneity of political connections, we ask respondents to identify which political party the firm's CEO supports. Using this question, we generate an indicator variable equal to 1 when the CEO supports the government party. We interact this variable with our political connections indicators. Given H2, we expect firms to benefit from political connections when CEOs are associated with government parties.


%We use several additional variables as controls. We control for \textit{firm size} and use a categorical measure of the number of employees. We also control for the firm's \textit{years in business}. We also include a binary indicator for \textit{foreign firms}, wholly owned or joint ventures. Lastly, we include two indicators equal to 1 for \textit{Ukrainian} or \textit{Egyptian} respondents, and 0 otherwise. We estimate a series of OLS models with robust standard errors and sector fixed effects. Table 1 presents results using perceptual political connections as the IV, while models in Table 2 use the count of officials and parliament members as the IV.

%\subsection{Ethical Considerations}

%As our study involved human subjects research, we first obtained approval for our study from our universities' Internal Review Boards (IRB). Two separate IRBs determined our research to be exempt of further review because our survey had less than minimal risk due to the fact that it focused on respondent perceptions of their companies and countries rather than on personal information. To complete the survey, all respondents had to complete a consent form which was also reviewed by the IRBs.

%To recruit respondents, we offered the equivalent of \$1 USD in mobile credit if they completed the survey. Because the dominant form of mobile plans in these countries involve purchasing mobile credit in small increments, the smaller denomination was still valuable as it provided a considerable amount of text, talk or data. Considering that the survey took no more than 20 minutes to complete, we believe it to be a sufficient reward for participation.\footnote{To provide context, the minimum wage in Egypt is approximately \$115 per month, \$3.50 per hour in Venezuela, and approximately \$2,000 per month in the Ukraine.} Furthermore, the incentive ensured that any mobile data used to complete the survey was reimbursed to the respondent.  

%We note that not all respondents chose to receive the credit, and not all respondents who submitted for the credit could be reimbursed because they did not enter their mobile numbers correctly. However, the majority of respondents received a credit. 


%\subsection{Experimental Treatment}\label{sec:experiment}

For these reasons, the experimental part of the survey involved a conjoint experiment that asked respondents to evaluate a series of hypothetical investment scenarios in which they choose how to allocate a \$100 investment between two companies. The respondents had some liberty in terms of how they interpreted the scenario in that we did not specify whether they are a representative of their company, their family or operating in a purely personal capacity given that these roles can be very difficult to separate in the countries we are studying \citep{almeida_theory_2005,ge_who_2019,kalhor_innovation_2020,samara_family_2020}. Because the precise make-up of the sources of capital of the potential investments is not the independent variable we wish to study, and difficult in any case to categorize in practice, we abstracted away from these details in order to maximize our realism in other aspects of the experimental design \citep{brutger_abstraction_nodate}. 

To ensure as much realism as possible with respect to the investment, the conjoint presented a variety of information on firms' characteristics, including political ties. In theory, if political connections are valuable to the firm--as much of the existing literature holds to be true in general--then we should expect more connected firms to receive more investment. By including numerous other attributes, we can make the scenarios much more compelling while also learning about the effects of a variety of criteria on investment decisions. \footnote{while we have tried our best to instill a high degree of realism in our experimental setup, we acknowledge the inherent limitations associated with any simulation. It is crucial to consider that actual investor behavior in real-world settings may deviate from the responses observed within our experimental context. As such, we recommend treating the experiment's results as indicative tendencies rather than absolute predictions of actual investor actions.} The presentation here closely follows that of our pre-registration in the supplementary information.

\begin{figure}[!h]
\caption{Conjoint Experiment}
	\centering
		\includegraphics[scale=.9]{Conjoint.png}
	\label{conjoint}
\end{figure}

In the experiment, respondents view two randomly generated company profiles paired together and are asked to choose some investment amount $x\in[0,100]$ to invest in one firm versus the other. Respondents evaluate four sets of investment pairs (i.e., tasks) and consequently assign investment amounts to 8 hypothetical firms. We present one example of a pair of randomly generated profiles in Figure \ref{conjoint}. This version of the figure shows the time-varying profitability and sales of the firm as parallel plots. We believe that this visual depiction communicates more realism about the hypothetical companies, though possibly also produces an additional informational burden on the respondent. As we discuss, for this reason we randomly substituted a Likert profitability scale in 50\% of the profiles.

%Our experiment has the format of a choice-based conjoint design between two alternatives, except that in this case, the respondent is choosing how to allocate a \$100 investment between two companies. The amount invested is a somewhat arbitrary number and we use \$100 because it is mathematically simple to divide. We have respondents choose between two options because it reduces satisficing: respondents need to consider different options and choose which one is best rather than opting for the first they see on the screen \citep{hainmueller2015}.\footnote{Although we ask individual to complete these tasks, it is reasonable to expect that individuals would act similarly when making investment decisions on behalf of their firm. This is especially likely to be the case for manager respondents. Since our results for the manager-only sample show a very high degree of similarity in responses compared to the full sample, this suggests that both managers and non-managers are evaluating prospective firms similarly, and would thus also make similar choice if they were making firm-level investment decisions.} In addition, the forced-choice design ensures that investment happens. If we only presented a single option, risk-averse respondents may never choose to invest at all. 

The possible attributes for this experiment include company ownership, company country of origin, sector, number of employees, current assets (\$), total sales (\$), total profit (\$), age of firm, and political connections. We sought to make the experiment as realistic as possible by generating plausible corporate data. For the economic attributes of the firm, we simulate from continuous distributions to approximate realistic investment decisions. We first draw the number of employees from series of uniform distributions that take 20\% of samples from the following bounds: (10 - 50), (50 - 100), (100 - 500), (500 - 1000) and (1000 - 5000). This produces a distribution with substantial mass in medium and small firms, with relatively fewer larger firms (greater than 1,000 employees). We draw the amount of capital as dollars per employee from a uniform distribution with a minimum of \$100 and a maximum of \$10000, and we draw a number of years for firm age from a uniform distribution between 3 and 65. 

%Given all these attributes is important to note that the respondent could well see company profiles which would be unlikely to be observed in their country. However, as our primary inference problem is sample selection bias, we note that randomization requires the respondent to see options which they would not normally otherwise see so that we can learn about these counterfactuals.


One of the challenges in designing a realistic investment scenario is to have data that an investor might expect about firm performance. Due to our uncertainty about how best to show data about a company, we take two approaches to manipulating corporate performances. First, we draw from a simple Likert list of attributes for firm profitability that run from Losing a Lot of Money to Very Profitable (5 categories). This approach uses heuristics to reduce information load for the respondent when evaluating treatment profiles at the risk of reduced realism for the profiles.

Our second approach was to provide the respondent with much more detailed randomized information about firm performance. To accomplish this, we generated individualized time series for each investment profile, which we then presented as time series graphs to the respondent. To generate the series, we drew time-varying total factor productivity $a_{it}$ value for each treatment. Conditional on these draws, we then calculated the amount of sales $q_{it}$ and profit $p_{it}$ for firm $i$ in year $t$ through a Cobb-Douglas production function given fixed values for labor $L_i$, capital $K_i$, and time-varying TFP $a_{it}$:

\begin{align*}
    q_{it} &= a_{it}\sqrt{L_{i}}\sqrt{C_{i}}\\
    p_{it} &= q_{it} - \sqrt{L_i}\sqrt{C_{i}}
\end{align*}

We used the square root of capital and labor to ensure constant returns to scale. Profit $p_{it}$ is equal to the amount of sales minus the same production function with unit TFP. In other words, TFP determines the amount of profit a firm receives. 

To allow TFP to vary over the lifetime of the firm, we drew two parameters $C_1$ and $C_2$ and generated TFP $a'_{it}$ via a quadratic function on the real line as a function of time $t$: 

\begin{align*}
    a'_{it} &= \alpha + C_1 a_{it} t + C_2 a_{it} t^2\\
    a_{it} &= e^{a'_{it}}
\end{align*}

We exponentiated $a'_{it}$ to $a_{it}$ because TFP must be a strictly positive parameter. Using a quadratic function for TFP allows the performance of the firm to vary over time in relatively smooth shapes, either concave-up, concave-down, convex-up or convex-down. It is also possible to have concave-convex and convex-concave shapes over the span of the firm, which would constitute S-curves where company performance decreased for a period and then increased for a period. To illustrate what the resulting TFP values look like, we drew 1000 random variates for a 100-time point trend and plotted them in Figure 4 in the supplemental information. This figure shows that we can capture a wide variety of shapes in this distribution that are all plausible ways that productivity (and hence profitability) could change in a firm over time. The respondent was not shown the TFP values directly (as they are in principle unobservable), only the net profit above or below sales.

The potential drawback of this approach is that it increased the information load of the treatment, possibly making it more difficult for the respondent to evaluate the treatment profile and leading to satisficing. The advantage of this version of the treatment is that is much more realistic and detailed than the Likert scale. Because of our uncertainty about which version would produce the most valid results, we randomized equally whether the respondent received a Likert scale for profitability or a full time series plot. We report results from this experiment-within-experiment in the supplementary information. In brief, it appeared that the inclusion of the time series plot caused respondents to spend about four additional minutes evaluating the profiles, possibly signaling increased cognitive load but also a higher quality of responses.

The rest of the attributes have discrete values, which we show in Table \ref{treatments2}. We measure political connections in numerous ways to try to capture the variety of relationships a company could have with the state. For example, either an owner or a board member could be politically-connected, and this relationship could be one of professional acquaintance (i.e. a former minister or bureaucrat) or one of friendship (former classmate) or familial tie (son-in-law, daughter-in-law, etc). This diversity in treatment type allows us to marginalize over all these possible kinds of connections so that when we present aggregated results we can know with confidence that the results are not simply due to the particular political connection of a certain type. In addition, we can know with much more precision what kind of political connections tend to be most or least valuable.

\begin{table}
	\centering
	\caption{Treatment Profiles for Investment Conjoint Survey Experiment}\label{treatments2}
	
	
	\begin{tabular}{l p{4cm} p{10cm}}
		Number & Attribute & Values\\
		\midrule
		1 & Ownership & 100\% Foreign-Owned, Domestic Private, Domestic Public \\[1em]
		2 & Country of Origin & United States, Germany, Saudi Arabia, China, Russia, Brazil, South Korea, Japan, Saudi Arabia\\[1em]
		3 & Sector & Telecommunications, Construction, Manufacturing, Agriculture, Retail, Energy, Financial Services\\[1em]
		4 & Government Relations & Member of Parliament is on board of company, Head of ministry is on board of company, Mid-level bureaucrat is on board of company, Owner is a former member of parliament, Owner is a nephew/niece of prime minister, Owner is former general in the military, Owner is former officer in the police, Owner is former classmate to the President, Owner is married to the President's daughter/son, Owner is member of the President's political party, President's son/daughter is on the board of the company, Owner has no interest in politics\\
		\bottomrule
	\end{tabular}

\end{table}

% Due to the large number of political connections treatments, we collapse these treatments into 5 subcategories when conducting the analysis: family ties, parliament member, police/military, general government ties, and no connections. The family ties category includes all treatments involving family connections. Police/military and parliament member correspond to treatments involving police or military ties or parliament member serving on the board, respectively. General connections correspond to all remaining government ties.

%We note that the investors in our survey can be interpreted in two different ways, as either individuals or corporate investors. The experimental treatment permits both interpretations because in developing countries the dividing line between personal, familial and corporate finance is often quite blurry . For our purposes, these two types of investment are both theoretically important as they constitute important sources of funds to domestic firms in markets that are credit-constrained. We leave it to future research to identify how personal versus corporate or familial decision-making affects these issues. 

\section{Results}

We present the experimental results both in terms of dis-aggregated treatments (i.e., all the different types of political connections) along with aggregated results in which the treatment indicator is binary (connected or unconnected). We use the binary indicator to see if treatment interactions tend to vary as predicted by our theory. Following our preregistration, we model the outcome using beta regression as the outcome is a bounded variable between 0 and 1. Because we have observations equal to 0 or 1 (i.e. all of the investment went to a particular company),  we use the ordered beta regression estimator \citep{kubinec_ordered_2022} to include these bounds. The results are presented as logit coefficients and have a similar interpretation to the more commonly used OLS-based conjoint estimator \citep{hainmueller2014} as differences in the proportion of funds invested in a given company relative to the baseline trait.\footnote{OLS results are similar for most models though we present these results as the distribution of the error term is more realistic for a DV with upper and lower bounds. In other words, we do not think that it makes sense to use a model that would allow investors to put more than 100\% of their capital into a company.} For reference, we included OLS estimation of the main results in the supplementary information, which shows minimal differences in either size or significance.

Figure \ref{allcon} shows all of the possible versions of political connections disaggregated in the standard form for presenting conjoint results, while Figure \ref{base} shows the rest of the conjoint survey attributes along with a collapsed binary indicator for whether a company had any type of political connections. As can be seen, while there are a wide varieties of ways that we coded political connections, the results show that the effect of connections is always negative. While some effects are larger than others--generally speaking, having an owner who is politically connected appears to be worse than having a board member who is politically connected--on the whole, there is considerable uniformity in how respondents evaluated political connections: they did not like them. This result went against what we believe to be the dominant message in the literature that political connections help companies secure higher returns and thus should be more attractive vehicles for investment. 

However, as we expressed in the previous section, there are serious inference problems to overcome in estimating the effect of political connections that is free of the influence of long-term selection biases, and for that reason it is perhaps not as surprising that our estimates depart from the those based on observational data. We interpret this result as implying there is still much we do not know how political connections affects economic decision-making, and we provide in this paper what we believe are the best explanations for this finding.
\begin{figure}
\centering
\includegraphics[width=0.9\linewidth]{all_con.png}
\caption{Dis-aggregated Political Connection Treatments}\label{allcon}
\end{figure}

In Figure \ref{base} we show the rest of the company attributes and their effects on investment. The plot shows that the factors that seemed to influence respondents most strongly were profitability and country of origin. Unsurprisingly, respondents invested the most in companies that were the most profitable. To produce a single profitability measure, we combined the continuous and Likert measures for profitability by collapsing the time series into categories based on concavity, i.e., convex-up profit series were considered the most profitable, and concave-down time series were considered to the least profitable. 

The results in Figure \ref{allcon} are consistent with the theoretical logic developed above. Respondents appear to have a more negative response to connections involving more influential political figures. This is consistent with the notion that ties to more powerful politicians imply fewer constraints and a higher risk of opportunism. Interestingly, companies originally from South Korea and Germany were selected as being more attractive for investment. By comparison, Russia and Venezuela were disfavored. As Venezuela was only shown to Venezuelean respondents, we can interpret the relationship as indicating a preference against domestic firms for Venezuelans, which follows from the level of instability in the country. However, these treatment effects are imprecisely estimated, and so we cannot rule out the possibility of a null effect or even a small probability of a true effect in the opposite direction.

We note as well that several of the factors did not seem to have much of an effect on investments, at least for the sample as a whole. The age of the firm, the type of firm (i.e. state-owned versus private), and the sector of the firm were not strong predictors of investment decisions. The exception are firms in the retail sector, which were relatively disfavored. Also, we note that respondents did not prefer companies with larger amounts of assets, though as we lack a clear theory to explain this result, we do not further comment on it.
\begin{figure}
\centering
\includegraphics[width=0.7\linewidth]{base_mod.png}
\caption{All Attributes in Conjoint Experiment}\label{base}
\end{figure}

%In the supplemental information we re-estimated these same results except with a sample of only respondents with managerial positions (see Supplemental Information Figures 1 and 2). We also re-estimated the models in which we exclude reported firm sectors that may not map on to the conventional understanding of firms, such as those that reported working in public administration or as a household business. We do not find noticeable differences that change our substantive conclusions--the effect of political connections is negative across all samples. What we do find is that the treatment effect is attenuated in our full sample relative to the manager sample by about 30\%. However, the manager-only sample's estimate is much more uncertain with an interval almost twice as wide. For these reasons, we report and interpret the results from our full sample as the bias-variance tradeoff would suggest that the evidence of bias is minimal while the additional variance (and consequently risk of making false inferences) we would gain by using the manager-only sample is quite high.

One aspect that is notable in Figures \ref{allcon} and \ref{base} are that the uncertainty intervals for some coefficients are very large in magnitude. This suggests that there is considerable heterogeneity in how respondents are evaluating companies. For this reason we turn to examining treatment effect heterogeneity. We note that although the specific analyses were perform were not pre-registered, they follow logically from our theory suggesting that political connections and other country and company-level attributes should affect how effective the treatment is.

We examine treatment effect heterogeneity by the respondent’s country and the political connectedness of the respondent’s company. Figure \ref{countryint} shows predicted values for investment decisions for treatment (connected) and control (unconnected) profiles by country (connections are coded as a binary variable). As can be seen, the treatment group receives less investment than the control group across all three countries, suggesting uniformity in how the treatment operates, though we note that the effect is noticeably larger in Venezuela than other countries. This result may be generated by differences across regimes in the sample. We return to these potential differences in the discussion by examining ways in which Venezuelan companies may have a different relationship vis-a-vis the state relative to businesses in Egypt and the Ukraine.

%Firms with political connections may encounter fewer constraints in Venezuela due to specific attributes inherent to its political system, such as increased predation, centralized regime control, and robust regime stability. Venezuela's high Corruption Perceptions Index (CPI) score of 16 and a low ranking on the Ease of Doing Business Index (188th out of 190) contrast sharply with Egypt's and Ukraine's better standings. For instance, Egypt scored 35 on the CPI and ranked 114th in the Ease of Doing Business Index, while Ukraine scored 30 on the CPI and ranked 64th on the Ease of Doing Business Index. In the challenging environment of Venezuela, politically connected firms might have access to a broader spectrum of resources and opportunities and enjoy regulatory leniency. These benefits could provide such firms with the resilience needed to withstand economic and political crises more effectively. As these are predictions with other variables held at their means, overlapping uncertainty intervals does not necessarily mean the underlying effects are statistically insignificant. 

\begin{figure}
\centering
\includegraphics[width=0.7\linewidth]{country_int.png}
\caption{Connections Treatment Subset by Respondent Country}\label{countryint}
\end{figure}

 Figure \ref{compareall} tests for whether issues with the sampling frame may affect treatment heterogeneity. We compare the full sample ATEs for political connections to a sample excluding possibly error-prone sectors (such as Other and Public Administration)\footnote{We do not remove these sectors as that would be a post hoc adjustment we did not pre-register and because we cannot be sure what respondents intended by reporting this sector. It could be, for example, that companies in the Public Administration sector are primarily government contractors, or that the respondent mis-reported the sector of their company} and a sample with only managers. Figure \ref{compareall} shows that the treatment effects are larger for the manager-only sample and somewhat larger for the sample with some sectors removed; this finding suggests that the full sample has some measurement error that attenuates the ATE. However, as the difference in effect sizes is relatively small while the difference in the variance of the estimates is quite large, we believe that our full sample results are the most reliable from a statistical point of view. We report additional analyses with changing sample definitions in section 4 of the SI, but these likewise show minimal differences in treatment effects as a function of the sample frame.
\begin{figure}
\centering
\includegraphics[width=0.7\linewidth]{compare_samples_all.png}
\caption{Comparison of Samples: Dis-aggregated Political Connection Treatments}\label{compareall}
\end{figure}

We report TFP and profitability treatment heterogeneity analysis in Figure 3 of the SI that shows our respondents evaluate connected and unconnected firms differently depending on whether they are profitable versus productive. Panel A in Figure 3 reveals that rising productivity does not cause rising investment proportions among connected firms, while panel B shows that rising profitability does cause increasing investment in connected firms. We interpret this discrepancy to suggest that respondents are aware that connected firms can tradeoff productivity for access to rents that would allow them to be very profitable despite having low productivity. Furthermore, excessive profitability could be an indication of access to lucrative connections. We note that our experimental design allows us to separate these effects; it would be very difficult to identify these different factors using observational data as TFP and profitability are endogenous to each other. 

We also examine whether a respondent's companies' political connections moderates the relationship between political connections as a treatment and investment decisions. To do so, we interact our two measures of respondent political connections, the 1-10 scores for total connections and the efficacy of connections, with a binary variable for connected vs. non-connected treatment profiles. The results are shown in Figure \ref{intcon}. Panel A in the figure shows the political connections score and panel B the efficacy of those connections, but as can be seen, the results are nearly identical. In either case, as a respondents' companies' connections increase, the probability of selecting a company for investment converges to the control distribution. In other words, respondents at companies that have high political connections that are helping their company do not penalize treatment profiles with connections.

\begin{figure}
\centering
\includegraphics[width=0.7\linewidth]{pol_con_int.png}
\caption{Interaction of Connections and Respondent Political Connection Variables}\label{intcon}
\end{figure}

The clear nature of this relationship suggests that respondents' companies' political connections is a strong moderator of how and whether they penalize a company for being politically-connected. Indeed, this interaction also helps us explain why the aggregate effect of connections is negative for the sample as a whole: because most respondents work at non-connected companies, they are subsequently less likely to select companies for investment that are connected. If our sample was composed entirely of respondents at politically-connected companies, we would likely to observe either a null or positive relationship between connections and investment.

While this subgroup relationship sheds light on the negative finding for the relationship between connections and investment, it is still not entirely clear why a respondent's company's connections should matter so much. We believe that this association is explained by similarities in how politically-connected companies operate which encourage cooperation. In particular, we know based on other questions in the survey that politically-connected companies are more likely to use informal rather than formal means of dispute resolution. These back-door, unofficial dispute resolution mechanisms could well dissuade potential investors who fear losing their investment to unaccountable company managers. To test this, we interact the binary treatment with the same count variable for informal means of dispute resolution that we used earlier (i.e., we count all the times that a company reported it used informal means to resolve some problem). We interact this variable and report the results in Figure \ref{expropint}. 
\begin{figure}
    \centering
    \includegraphics[width=\linewidth]{exprop_treatment.png}
    \caption{Interaction of Political Connections Treatment and Informal Means of Enforcement}
    \label{expropint}
\end{figure}
While the relationship in Figure \ref{expropint} is less precise than the political connections interaction, we can identify a very similar pattern: respondents at companies that are more likely to use informal means of dispute resolution are less likely to penalize companies with political connections as potential investment vehicles. The limitation in terms of uncertainty at the top end of the scale is largely an artifact of the sample because only a minority of our respondents are at politically-connected companies that use informal means of dispute resolution. As a result, the relationship is much more precise for the larger number of companies that do not use informal means of dispute resolution.

In the supplemental information we also perform a mediation analysis in which we let the respondent's political connections predict the number of times informal dispute resolution is used. While there is a strong relationship between these two variables, the net result is inconsistent mediation, making it difficult to know to what extent the number of informal dispute resolution may be a mediator for a respondent's political connections. We do though find a strong positive direct effect from dispute resolution on investing in connected companies, suggesting that when we disentangle the causal sequence, informal dispute resolution is even more strongly related to being willing to invest in connected companies  compared to what Figure \ref{expropint} shows.

%Unfortunately, this table does not provide an unambiguous answer to the question. While the coefficient is positive, suggesting that respondents at companies that practice informal means of dispute resolution are more likely to invest in companies with political connections, the result does not reach statistical significance. To clarify matters, we plot the conditional effect of political connections by varying levels of informal enforcement in Figure \ref{expropint}. This figure shows that at low levels of informal enforcement, the control group has significantly higher levels of investment than the connected group, but at higher levels of enforcement there is not enough statistical power to detect effects. As such, while we believe there to be some evidence of a relationship, it is not conclusive.

%In Model 3 we try to explore the more subtle nuances of profitability from our experimental treatments that allowed profitability to change over time as TFP varied. To do so, we fit a quadratic linear regression model to each treatment's profitability time series, and then recorded the linear and quadratic coefficient from the model. We then interacted these coefficients with the binary coding for whether profitability increased in total over the whole sample. These results show that the triple interaction between the linear term, the squared term and positive performance is statistically significant, though the other terms are not so. This model shows that the respondents were able to absorb at least some of the more subtle dynamics of the treatment, although we note it is relatively hard to capture those dynamics in a linear model in a way that is easy to interpret.
%Finally, in Table \ref{gender} we examine gendered differences among treatments by coding each political connection treatment for whether the mentioned connection had a feminine gender, a masculine gender or no revealed gender (non-gendered) in Model 1. The baseline category for this variable is the control condition. We also interacted the gender treatment with respondent's country (Model 2) and also company profitability as a 5-point Likert scale (Model 3). These results very intriguingly show that overall, respondents did not prefer companies with feminine gendered political connections (Model 1), and this penalty against feminine connections was more true in Egypt relative to Venezuela and the Ukraine. Furthermore, respondents tended to punish feminine connections for poor company performance more than they punished male political connections (Model 3). On the whole, these results suggest that respondents viewed masculine gendered political connections more favorably, and penalized masculine gendered connections less strongly for poor company performance.

\section{Discussion}

While this survey experiment has turned up intuitive findings that help us understand the dynamics of political economy in developing countries, it is important to note that survey experiments by definition have limited external validity. We cannot assume that investors would react the same way to an actual appeal for funds as they would to a hypothetical scenario. We attempted to make the experiment as realistic as possible by simulating plausible firm performance data, but invariably if real money were at stake, other issues would no doubt appear that we did not address. 

However, we believe that the treatment effects reported are still meaningful as our experimental design means that we know that confounding variables cannot explain differences between control and treatment groups. While the treatment was undoubtedly weaker than a real-world situation would be, the fact that respondents interpreted political connections in similar ways suggests that their responses are indicative of the institutional context for businesses in their country. These perceptions have been shaped by their experiences as businesspeople and reflect their best judgment about how they would use their money, and probably in many cases, have done so. 

We next return to the question of country-level heterogeneity. We note first that it was not the intent of the research design to investigate cross-country differences but rather to recruit a diverse sample of businesspeople, that is, to maximize the external validity of the research design. As such, our analysis here is purely exploratory; if we had not observed treatment heterogeneity at the country level we might not have looked further into cross-country differences. However, with this caveat, we believe there are important country-level factors that are worth considering.

\begin{figure}
    \centering
    \caption{Business-State Relations Across Countries}
    \label{fig:biz_state}
    \includegraphics[width=.9\linewidth]{venezuela_diff.pdf}
\end{figure}

As we reported in Figure \ref{countryint}, Venezuelan respondents show a stronger anti-political connections effect relative to other countries (though the effect is negative in all countries, if not as strong). We believe that the most plausible explanation for this discrepancy is that business-state relations between companies and the regime is significantly worse on the whole than in Ukraine or Egypt. While Ukraine and Egypt have notable issues with corruption, and neither can be considered a consolidated democracy, the regime is not as much at war with society as the Venezuelan regime is. 

We can illustrate this disparity by looking at other survey questions in Figure \ref{fig:biz_state}. Venezuelan companies are less likely to report having political connections (panel A), much more likely to report paying more than 30\% of their income in bribes (panel B), more likely to report increases in bribe payments (panel C), and much more likely to report that the political situation in their country has deteriorated (panel D). While political connections are surely still valuable in Venezuela--if for no other reason than to avoid having to pay such a large sum in bribes--the antagonism between the regime and society appears to affect businesspeople's willingness to invest in companies where the state has a presence. In other words, it would appear that the Venezuelan state has moved closer to the predatory state archetype than either Ukraine or Egypt.



\section{Conclusion}

In this paper, in order to better understand how people make investment decisions in countries with poor quality of institutions, we implement an online survey using Facebook advertising to recruit subjects who are either business employees or managers. We use finely grained information about political connections and experimental manipulations to be able to assess the role of political connections as opposed to other factors. In contrast to what we might expect given the value of political relationships, we find that political connections can be a barrier rather than an aid to domestic investment in developing countries.

On the whole, political connections discourage investment in companies in our experiment, but we find that such an effect is partly explained by the role of political connections at the respondent's company. Respondents in more connected companies and companies that benefited from connections were more likely to choose connected companies, and vice versa. This result strongly suggests that the benefits of political connections to a company are not unequivocal but depend on that company's position vis-a-vis those who are currently in power. 

For these reasons, we believe that political connections are a more risky strategy for companies than has often been acknowledged in the literature. At first glance, it seems counter-intuitive that stronger political connections can lead to more government predation or expropriation, and that investors do not seem to prefer politically-connected companies. However, these findings are driven by the highly conditional effects of connections. 



\label{sec:conclusion}

\newpage

% \subsubsection{Perceptual Measure of Political Predation}

% 1. Please evaluate the risk of expropriation of your business premises
% \vspace{5pt}
% \bigskip
% \\
% \begin{tabular}{ccccc}
% \centering
% \begin{tabular}{ccccc}
% Very Low & Low & Moderate & High & Very High \\
% ㅁ        & ㅁ   & ㅁ        & ㅁ    & ㅁ        
% \end{tabular}      
% \end{tabular}
% \bigskip
% \\
% 2. Please tell us to what extent your firm's managers would agree or disagree with the following statements:
% \begin{center}
% \resizebox{\textwidth}{!}{%
% \begin{tabular}{lccccc}
%                                                                                                                   & Strongly agree & Somewhat agree & Neither agree or disagree & Somewhat disagree & Strongly disagree \\
% \begin{tabular}[c]{@{}l@{}}Changes in regulatory standards are transparent, \\ predictable and fair.\end{tabular} & ㅁ              & ㅁ              & ㅁ                         & ㅁ                 & ㅁ                 \\
% The court system is fair, impartial and uncorrupted                                                               & ㅁ              & ㅁ              & ㅁ                         & ㅁ                 & ㅁ                 \\
% The police are fair, impartial and uncorrupted                                                                    & ㅁ              & ㅁ              & ㅁ                         & ㅁ                 & ㅁ                
% \end{tabular}%
% }
% \end{center}


% \bigskip
% 3. Would you say that the number of inspections your firm experienced from these agencies is higher or lower than average for your industry?

% \vspace{5pt}
% \bigskip

% \resizebox{\textwidth}{!}{%
% \begin{tabular}{lccccc}
%                       & Much Lower than Average & Lower than Average & Average & Higher than Average & Much Higher than average \\
% Tax Agency             & ㅁ & ㅁ& ㅁ& ㅁ& ㅁ                    \\
% Police/Law Enforcement & ㅁ & ㅁ& ㅁ& ㅁ& ㅁ                    \\
% Safety Inspectors      & ㅁ & ㅁ& ㅁ& ㅁ& ㅁ                    \\
% Health Inspectors     & ㅁ & ㅁ& ㅁ& ㅁ& ㅁ                   
% \end{tabular}
% }
% \\

%===================References=====================
\newpage
\singlespacing
\footnotesize
\bibliographystyle{apsr}
\bibliography{PandE.bib}

%===================End Document===================

\end{document}